Index: openmp/runtime/src/kmp_tasking.cpp
===================================================================
--- openmp/runtime/src/kmp_tasking.cpp
+++ openmp/runtime/src/kmp_tasking.cpp
@@ -3896,8 +3896,9 @@
 
   // We need to wait to make sure the top half is finished
   // Spinning here should be ok as this should happen quickly
-  while (KMP_ATOMIC_LD_ACQ(&taskdata->td_incomplete_child_tasks) > 0)
-    ;
+  if (!taskdata->td_flags.detachable)
+    while (KMP_ATOMIC_LD_ACQ(&taskdata->td_incomplete_child_tasks) > 0)
+      ;
 
   __kmp_release_deps(gtid, taskdata);
   __kmp_free_task_and_ancestors(gtid, taskdata, thread);
Index: openmp/runtime/test/tasking/detach_nested_task.c
===================================================================
--- /dev/null
+++ openmp/runtime/test/tasking/detach_nested_task.c
@@ -0,0 +1,54 @@
+// RUN: %libomp-compile-and-run
+
+// The outer detachable task creates multiple child tasks with dependencies
+// when the last inner task incremented ret, the task calls omp_fulfill_event
+// to release the outer task.
+
+#include <omp.h>
+#include <stdio.h>
+
+int *buf;
+
+int foo(int n)
+{
+  int ret = 0;
+  for (int i = 0; i < n; ++i) {
+    omp_event_handle_t event;
+    #pragma omp task detach(event) firstprivate(i,n,event) shared(ret) default(none)
+    {
+      for (int j = 0; j < n; ++j) {
+        #pragma omp task firstprivate(event,i,j,n) shared(ret) default(none) depend(out:ret)
+        {
+          //printf("Task %i, %i: %i\n", i, j, omp_get_thread_num());
+          #pragma omp atomic
+            ret++;
+#if _OPENMP
+          if (j == n-1) {
+            //printf("Task %i, %i: omp_fulfill_event()\n", i, j);
+            omp_fulfill_event(event);
+          }
+#endif
+        }
+      }
+    }
+  }
+  // the taskwait only guarantees the outer tasks to complete.
+  #pragma omp taskwait
+
+  return ret;
+}
+
+
+int main()
+{
+  int ret;
+#pragma omp parallel
+#pragma omp master
+  {
+    ret = foo(8);
+  }
+  printf("%i\n", ret);
+  //CHECK: 64
+  return 0;
+}
+